Assessment of Annual Rainfall Patterns Using a Clustering-based Comparative Framework

Recent decades have seen a dramatic increase in destructive rainfall events, leading to concerns regarding the uniqueness and severity of these events. The literature does not provide an accepted definition for “patterns” for assessing rainfall severity. Hence, the main challenge lies in utilizing flexible patterns that can effectively reflect variations within annual rainfall datasets and suitable for diverse seasons and climate contexts. This study aimed to address the severity of annual rainfall records and investigate how they vary between lenient and intense years in Wajima City, Japan. For this, a clustering-based framework was proposed. The framework comprises two key approaches: (i) pentad precipitation integrated with hierarchical clustering, and (ii) information-complexity analysis integrated with partitional clustering. Pentad analysis could provide a comprehensive image of key patterns at sub-monthly scales. As a result, it was found that heavy-rainfall incidents were intense mainly during the autumn rain front. Importantly, this study offered an innovative application of information and complexity analysis, which may provide a new perspective for the assessment of hydrological systems and diverse environmental systems. The findings demonstrated that information and complexity metrics can be used as deductive indicators to capture hidden patterns in annual rainfall incidents, such as the frequency and persistence of major rainfall events. As the mean information gain increased, the number of days that exceeded the daily rainfall threshold increased, and vice versa. In addition, a higher effective complexity was associated with more prolonged rain incidents. Overall, the findings provide an important reference for rainfall assessments.